Legal claims defining the scope of protection. Each claim is shown in both the original legal language and a plain English translation.
1. A computer-implemented method for presenting similar images on a display device, comprising: displaying a first image on the display device; determining one or more homographic relationships between the first image and a plurality of images; identifying, using the determined one or more homographic relationships, at least one image having a scene and a perspective which are similar to those of the first image; and displaying the identified image; wherein the determining one or more homographic relationships comprises: creating a homography graph from a collection of images, wherein respective images of the collection are nodes of the graph and two nodes of the graph have an edge between them if the two nodes have matching image features, and wherein a cost of the edge is a homography relationship between the two nodes; generating homography clusters for respective images of the collection of images using the created homography graph; and determining a homographic relationship between images of one of the generated homography clusters.
The computer system displays an initial image and finds similar images by calculating how the initial image relates to a collection of other images, displaying one or more similar images to the user. Specifically, the system creates a homography graph where each image is a node, and connections (edges) exist between images with matching visual features. The "cost" of each connection is the homography between those images. The system then groups the images into "homography clusters." The system then finds homographic relationships (geometric transformations) between images in one of these clusters.
2. The computer-implemented method of claim 1 , wherein the scene of the identified image is substantially the same as that of the first image, and wherein the perspective of the identified image is nearby to that of the first image.
Building upon the method for finding similar images, the system identifies similar images where the overall scene content is nearly identical to the initial image, and the viewing angle or perspective of the similar image is close to the initial image's perspective. In other words, the found image shows nearly the same objects in the same layout, viewed from a slightly shifted position.
3. The computer-implemented method of claim 2 , wherein the identifying includes: accessing a homography matrix associated with one of the plurality of images; and selecting the one of the images as the identified image wherein the accessed homography matrix is substantially an identity matrix.
Extending the method for finding similar images, the system accesses a homography matrix associated with one of the other images, and the system selects that "other image" as the similar image to be displayed if that matrix closely approximates an "identity matrix." (an identity matrix is one that causes no change or transformation). This means the selected image has almost no transformation from the first image.
4. The computer-implemented method of claim 3 , wherein the selecting is based upon a similarity threshold between the accessed homography matrix and the identity matrix.
Building upon the method of selecting similar images using a homography matrix, the system selects an image as similar only if the homography matrix associated with that image is similar enough to the identity matrix, as measured against a predefined similarity threshold. This allows setting the strictness of what is considered "the same" perspective.
5. The computer-implemented method of claim 1 , further comprising: receiving a user-selected geographic location coordinate; and selecting the first image based upon the received geographic location coordinate.
Further extending the method for finding similar images, the system can receive a geographic location coordinate from the user (e.g., GPS coordinates). The system then selects the initial image to display based on this geographic location. The subsequent process of homography-based image searching then starts from this geolocated initial image.
6. The computer-implemented method of claim 5 , wherein the selecting the first image comprises: identifying an image associated with a corresponding geographic location coordinate, wherein the image is included in a cluster from a plurality of homographic clusters, and wherein the plurality of homographic clusters is generated using a homography graph; and returning the identified image as the selected first image.
Further extending the method for finding similar images based on geographic location, the system identifies an image that is associated with geographic coordinates that match or are close to the user-selected geographic location. This image must also be part of a homography cluster, where the clusters were generated using a homography graph. The identified image is then returned as the selected first image, and used to find further similar images.
7. The computer-implemented method of claim 1 , further comprising: matching of the first image to respective images of a collection of images based upon image features to determine the plurality of images.
In addition to the method of finding similar images using homographies, the system matches the first image to other images in a collection, based on shared image features. This matching process generates the plurality of images to which the first image will be compared in the later homography steps.
8. The computer-implemented method of claim 1 , wherein the generating homography clusters comprises: selecting a node in the homography graph; propagating, recursively to each neighbor of the node, a representative area; and adding the neighbor to a homography cluster of the node if the propagated representative area is within a significant area of a corresponding image of the neighbor.
Regarding the method of generating homography clusters, the system selects an image (node) in the homography graph and then recursively propagates a representative area from that node to each of its neighbors in the graph. A neighbor is added to the homography cluster of the initial node if the propagated representative area is within a certain area of the neighbor's corresponding image, meaning the areas align geometrically.
9. The computer-implemented method of claim 8 , wherein the propagating of the representative area is performed by transforming the representative area using the homography relationship corresponding to the edge between the two nodes.
Expanding on the method for generating homography clusters, when propagating the representative area from one image to its neighbor in the homography graph, the system transforms the representative area using the homography relationship that exists between the two images (represented as the "cost" of the edge between the nodes).
10. The computer-implemented method of claim 1 , wherein respective ones of the generated homography clusters include images that are not immediate neighbors in the homography graph.
Within the method of generating homography clusters, each cluster can contain images that are *not* directly connected (immediate neighbors) to each other in the homography graph. This means that the clustering process can group images that are related but not directly visually similar.
11. A system for presenting similar images, comprising: a processor; a display device coupled to the processor and configured to display a first image; an image homography generator executed on the processor and configured to determine homographic relationships between the first image and a plurality of images; and a similar image selector executed on the processor and configured to identify, using the determined homographic relationships, at least one image having a scene and a perspective which are substantially similar to those of the first image; wherein the image homography generator is further configured to determine one or more homographic relationships by: creating a homography graph from a collection of images, wherein respective images of the collection are nodes of the graph and two nodes of the graph have an edge between them if the two nodes have matching image features, and wherein a cost of the edge is a homography relationship between the two nodes; generating homography clusters for respective images of the collection of images using the created homography graph; and determining a homographic relationship between images of one of the generated homography clusters.
The system displays an initial image and finds similar images by calculating homographic relationships between the initial image and a collection of other images and displays one or more similar images to the user. The system includes a processor, a display, an image homography generator, and a similar image selector. The homography generator creates a homography graph where each image is a node, and connections (edges) exist between images with matching visual features. The "cost" of each connection is the homography between those images. The system groups the images into "homography clusters," and finds homographic relationships between images in one of these clusters.
12. The system of claim 11 , wherein the similar image selector is further configured to identify the at least one image, wherein the scene of the identified image is substantially the same as that of the first image, and wherein the perspective of the identified image is nearby to that of the first image.
In the system for finding similar images, the similar image selector identifies images where the overall scene content is nearly identical to the initial image, and the viewing angle or perspective of the similar image is close to the initial image's perspective.
13. The system of claim 11 , wherein the similar image selector is further configured to: access a homography matrix associated with the first image and one of the plurality of images; and select the one of the images as the identified image wherein the accessed homography matrix is substantially an identity matrix.
In the system for finding similar images, the similar image selector accesses a homography matrix associated with the initial image and another image in the collection. The selector then selects the other image as similar if its associated homography matrix closely approximates an identity matrix (one that causes no change).
14. The system of claim 13 , wherein the selecting is based upon a similarity threshold between the homography matrix and the identity matrix.
In the system for finding similar images, the selector chooses an image as similar only if the homography matrix is close enough to the identity matrix, as measured by a defined similarity threshold.
15. The system of claim 11 , further comprising an initial image picker executed on the processor and configured to: receive a user-selected geographic location coordinate; and select the first image based upon the received geographic location coordinate.
The system for finding similar images based on homographies also includes an initial image picker. This image picker receives a geographic location from the user and selects the first image based on this location.
16. The system of claim 11 , further comprising a matched image cluster generator executed on the processor and configured to match the first image to respective images of a collection of images based upon image features to generate the plurality of images.
The system for finding similar images also includes a "matched image cluster generator." This component matches the initial image to other images in the collection based on image features. This matching generates the set of images to which the first image will be compared using homographies.
17. An article of manufacture comprising a non-transitory computer readable medium encoding instructions thereon that in response to execution by a computing device cause the computing device to perform operations comprising: displaying a first image on the display device; determining homographic relationships between the first image and a plurality of images; identifying, using the determined homographic relationships, at least one image having a scene and a perspective which are substantially similar to that of the first image; and displaying the identified image; wherein the determining homographic relationships comprises: creating a homography graph from a collection of images, wherein respective images of the collection are nodes of the graph and two nodes of the graph have an edge between them if the two nodes have matching image features, and wherein a cost of the edge is a homography relationship between the two nodes; generating homography clusters for respective images of the collection of images using the created homography graph; and determining a homographic relationship between images of one of the generated homography clusters.
A computer-readable medium contains instructions that, when executed, cause the computer to display an initial image and find similar images by calculating how the initial image relates geometrically to a collection of other images. The process displays one or more similar images to the user. The system creates a homography graph where each image is a node, and connections exist between images with matching visual features. The system groups the images into "homography clusters," and finds homographic relationships between images in one of these clusters.
18. The article of manufacture of claim 17 , wherein the identifying includes: accessing a homography matrix associated with one of the plurality of images; and selecting the one of the images as the identified image wherein the accessed homography matrix is substantially an identity matrix.
In this computer-readable medium, the step of identifying similar images includes accessing a homography matrix associated with one of the other images, and selecting that other image as the similar image to be displayed if that matrix closely approximates an "identity matrix" (meaning the image does not change).
Unknown
November 11, 2014
Browse 5M+ US patents with plain-English claim translations and AI-generated analysis.